Fault detection effectiveness of spathic test data

This paper presents an approach for generating test data for unit-level, and possibly integration-level, testing based on sampling over intervals of the input probability distribution, i.e., one that has been divided or layered according to criteria. Our approach is termed "spathic" as it selects random values felt to be most likely or least likely to occur from a segmented input probability distribution. Also, it allows the layers to be further segmented if additional test data is required later in the test cycle. The spathic approach finds a middle ground between the more difficult to achieve adequacy criteria and random test data generation, and requires less effort on the part of the tester. It can be viewed as guided random testing, with the tester specifying some information about expected input. The spathic test data generation approach can be used to augment "intelligent" manual unit-level testing. An initial case study suggests that spathic test sets defect more faults than random test data sets, and achieve higher levels of statement and branch coverage.